<p><strong>Location:</strong> Munich, Germany</p><p><strong>Language:</strong> German (B2+ required) & English</p><p><strong>Work Model: </strong>Hybrid (On-site hardware lab days)</p><p><strong>Position Type:</strong> Full-time, Permanent</p><p><br></p><p><strong><u>The Opportunity</u></strong></p><p>On behalf of a highly innovative, well-funded pioneer in the German robotics ecosystem, we are seeking a visionary Generative AI / Large Behaviour Model (LBM) Specialist. Operating at the absolute frontier of Physical AI, this company is actively moving away from traditional, deterministic programming. Instead, they are building cutting-edge neural architectures designed to give intelligent physical agents the ability to perceive, reason, and autonomously execute complex manipulation tasks in unstructured environments.</p><p><br></p><p>Based in Munich, this organization combines the fast-paced agility of an elite deep-tech research lab with the robust engineering scaling infrastructure of an established industry disruptor. This position offers a rare hybrid balance: the flexibility of remote software development paired with critical hands-on hardware validation in a world-class robotics laboratory.</p><p><br></p><p><strong>The Core Mission:</strong> Transitioning advanced generative models from the digital realm to physical reality. The successful candidate will spearhead the development of multi-modal architectures where high-level semantic intent seamlessly translates into zero-shot physical execution.</p><p><br></p><p><strong><u>Key Responsibilities</u></strong></p><ul><li>Architecture Development: Design, train, and fine-tune Vision-Language-Action (VLA) models and large behaviour Models (LBMs) tailored for low-latency, multi-modal robot control.</li><li>Sim-to-Real Pipeline Optimization: Build and scale scalable training pipelines utilizing hyper-realistic simulation physics environments to train generative policies at scale before deploying models to physical hardware.</li><li>End-to-End Learning Systems: Develop, test, and optimize end-to-end neural networks capable of directly mapping raw multi-modal sensor inputs (visual, tactile, spatial telemetry) into real-time, fluid motor commands.</li><li>Cross-Functional Hardware Integration: Collaborate closely with Embedded Software Engineers and Control System Engineers to deploy, benchmark, and iterate on models running directly on edge compute units (e.g., NVIDIA Jetson platforms).</li><li>State-of-the-Art Research Transition: Monitor, replicate, and adapt the latest machine learning breakthroughs (Diffusion Policies, Transformer-based architectures, and generative behaviour cloning) from top academic research into production-ready commercial frameworks.</li></ul><p><br></p><p><strong><u>Candidate Profile</u></strong></p><p>The ideal candidate sits at the intersections of modern deep learning theory, generative AI architecture, and physical hardware deployment. They are inherently curious, analytical, and possess a deep passion for solving the unstructured, real-world edge cases of Embodied AI.</p><p><br></p><p><strong><u>Required Qualifications & Skills</u></strong></p><ul><li>Education: Master’s degree or PhD in Computer Science, Robotics, Machine Learning, Physics, or a highly quantitative field with a specialized focus on deep learning.</li><li>Generative AI & Architecture Experience: Proven track record of developing and scaling Transformers, Diffusion Models, or large multi-modal foundational models. Specific experience adapting these models for spatial reasoning, navigation, or physical actions is highly advantageous.</li><li>Core Tech Stack: Exceptional production-grade programming skills in Python and deep familiarity with machine learning frameworks such as PyTorch or JAX.</li><li>Robotics Frameworks: Direct, hands-on experience or profound theoretical understanding of ROS2 (Robot Operating System), data serialization, and robotic kinematics.</li><li>Simulation Tools: Prior experience with modern physics engines and simulation environments (e.g.,NVIDIA Isaac Sim, MuJoCo, Drake) for robot learning.</li></ul>